SINCE orthogonal frequency division multiplexing

IEEE TRANSACTIONS ON BROADCASTING, VOL. 55, NO. 3, SEPTEMBER 2009 679 Bit-Based SLM Schemes for PAPR Reduction in QAM Modulated OFDM Signals Hyun-Bae Jeon, Kyu-Hong Kim, Jong-Seon No, Member, IEEE, and Dong-Joon Shin, Member, IEEE Abstract In this paper, we propose two bit-based selected mapping (SLM) schemes for reducing peak to average power ratio (PAPR) of orthogonal frequency division multiplexing (OFDM) signals with quadrature amplitude modulation (QAM), called bitwise SLM (BSLM) and partial bit inverted SLM (PBISLM). Contrary to the conventional SLM which rotates the phases of QAM symbols in the frequency domain, the proposed schemes change the magnitudes as well as the phases of QAM symbols by applying binary phase sequences to the binary data sequence before mapped to QAM symbols. Simulation results show that the proposed schemes have better PAPR reduction performance with shaping gain than the conventional SLM scheme for the QAM modulated OFDM signals, especially for the small number of subcarriers. Index Terms Orthogonal frequency division multiplexing (OFDM), peak to average power ratio (PAPR), quadrature amplitude modulation (QAM), selected mapping (SLM), shaping gain. I. INTRODUCTION SINCE orthogonal frequency division multiplexing (OFDM) can support high data rate and provide high reliability in voice, data, and multimedia communications, it has been adopted as a standard technique in many wireless communication systems. One of main advantages of OFDM is the robustness against frequency selective fading or narrowband interference. However, a critical drawback of OFDM is high peak to average power ratio (PAPR) which results in significant inter-modulation and undesirable out-of-band radiation when an OFDM signal passes through nonlinear devices such as high power amplifier (HPA) [1]. Several techniques have been proposed to mitigate the PAPR of OFDM signals. Clipping is used to reduce the peak power by clipping the OFDM signals to the threshold level [2] but it causes inband distortion and out-of-band radiation. Companding schemes scale the time-domain signals nonlinearly such that the signals with large amplitude are suppressed and the signals with small amplitude are expanded, which also distorts the signals unavoidably [3] [5]. Tone reservation (TR), tone injection (TI) [6], and active constellation extension (ACE) [7] utilize some subcarriers only to reduce PAPR. The Manuscript received February 26, 2009; revised May 05, 2009. First published July 17, 2009; current version published August 21, 2009. This work was supported in part by the Korea Science and Engineering Foundation (KOSEF) grant funded by the Korea government(mest) (No. 2009-0081441). H.-B. Jeon, K.-H. Kim, and J.-S. No are with the Department of Electrical Engineering and Computer Science, INMC, Seoul National University, Seoul 151-744, Korea, (email: lucidream@ccl.snu.ac.kr; adageneral@ccl.snu.ac.kr; jsno@snu.ac.kr). D.-J. Shin is with the Department of Electronics and Computer Engineering, Hanyang University, Seoul 133-791, Korea (e-mail: djshin@hanyang.ac.kr). Digital Object Identifier 10.1109/TBC.2009.2025977 drawbacks of these methods are data rate loss or transmission power increment. Selected mapping (SLM) [8] and partial transmit sequence (PTS) [9] select the signal with the minimum PAPR among several candidate signals generated by multiplying phase sequences to the data sequence before or after inverse fast Fourier transform (IFFT). To overcome the high computational complexity that these schemes accompany inherently, several SLM and PTS schemes have been proposed to improve the PAPR reduction performance and reduce the computational complexity [10] [13]. Also, some SLM schemes such as bit interleaving or scrambling [14], [15] modify binary data sequence before applying quadrature amplitude modulation (QAM) to generate alternative symbol sequences. In these cases, alternative symbol sequences undergo the change of both amplitude (or power) and phase. In this paper, we reconsider the conditions on the phase sequences for good SLM scheme and analyze the relation between the independency of alternative signal sequences and the covariance of the average symbol powers of them. Based on these results, we propose two bit-based SLM schemes which vary not only the phase but also the power of data symbols as follows. Bitwise SLM (BSLM) generates the alternative symbol sequences through multiplying a binary data sequence by binary phase sequences with the same length as that of the binary data sequence. Partial bit inverted SLM (PBISLM) generates the alternative symbol sequences by inverting data bits at predetermined bit positions or not according to the binary phase sequences. The PAPR reduction performance of the proposed SLM schemes is better than that of the conventional SLM scheme for the QAM modulated OFDM signals, especially for the small number of subcarriers. Moreover, they can reduce the average transmission power compared to the conventional SLM, which is called the shaping gain. The rest of this paper is organized as follows. In Section II, OFDM system and conventional SLM scheme are reviewed. The conditions on the phase sequences for good SLM scheme are derived in Section III and two bit-based SLM schemes are proposed in Section IV. The numerical analysis is provided in Section V and the conclusions are given in Section VI. II. CONVENTIONAL SLM SCHEME An OFDM signal sequence subcarriers can be expressed as using where is an input symbol sequence usually modulated by using phase shift keying (PSK) or QAM and (1) 0018-9316/$26.00 2009 IEEE

680 IEEE TRANSACTIONS ON BROADCASTING, VOL. 55, NO. 3, SEPTEMBER 2009 stands for a discrete time index. The PAPR of the OFDM signal sequence in the discrete time domain can be defined as where denotes the expectation operator. In the conventional SLM scheme, a transmitter generates distinct alternative symbol sequences, all representing the same input symbol sequence, and selects the one with the minimum PAPR for transmission. To generate alternative symbol sequences, an input symbol sequence is multiplied by different phase sequences of length,,,. The first phase sequence is usually the all-1 sequence. Then, the alternative symbol sequences,, are generated by. After alternative symbol sequences are transformed by IFFT, the alternative OFDM signal sequence with the smallest PAPR is selected for transmission. If we assume that the alternative OFDM signal sequences,, are mutually independent, the complementary cumulative distribution function (CCDF) of PAPR for the SLM scheme can be given [9] as (3) There are some design criteria for good phase sequences of the conventional SLM scheme. In [16], two criteria are suggested, one is the orthogonality between phase sequences and the other is the aperiodicity of the phase sequences. It is shown in [17] that the alternative OFDM signal sequences are asymptotically mutually independent if the phases of symbols in each phase sequence are independent and identically distributed (i.i.d.) with zero expectation value, and the SLM scheme satisfying this condition can have the optimal PAPR reduction performance. III. CONDITIONS FOR MUTUALLY INDEPENDENT OFDM SIGNALS OFDM signal sequences obtained after IFFT can be assumed to be complex Gaussian distributed for large by the central limit theorem. Thus zero covariance of two alternative OFDM signals implies that they are mutually independent [17]. However, this assumption does not hold for small because OFDM signal sequence for small may not be approximated as complex Gaussian random vector. It is known that the CCDF of PAPR for the conventional SLM scheme and the theoretical CCDF are almost identical for QPSK, but they become different for 16-QAM even though the phase sequences satisfy the optimal conditions in [16], [16]. Also, this difference becomes larger as increases or decreases. These observations lead us to consider new SLM schemes for OFDM symbol with non-constant envelope modulation. If OFDM signal sequences are not complex Gaussian distributed, zero covariance of two alternative OFDM signals does (2) not guarantee the mutual independency between them. In this case, instead of covariance, we consider the property of joint cumulants of alternative OFDM signals such that two alternative OFDM signal sequences are mutually independent if the joint cumulants of all orders are equal to zero [19]. Since it is not easy to calculate high order joint cumulants, we will only consider joint cumulants up to the fourth order to investigate the independency of alternative OFDM signal sequences. In general, the th symbol of the th alternative symbol sequence can be expressed as (4) where and are the amplitude gain and phase rotation of the th symbol in the th alternative symbol sequence, respectively. Note that in the conventional SLM, for all and. Then the th alternative OFDM signal sequence is given as It can be easily shown that the second and third order joint cumulants are zero regardless of if is i.i.d. with for. Thus we will propose new SLM schemes which make the fourth order joint cumulants of any pair of alternative OFDM signals close to zero. Through numerical analysis, it will be shown that the PAPR reduction performance improves as the fourth order joint cumulant between alternative OFDM signal sequences decreases. If the phase sequences satisfy the i.i.d. and mean zero conditions for the phase, the fourth order joint cumulant of two alternative OFDM signal sequences can be given as It can be easily shown that the fourth order joint cumulant in (6) is equivalent to the covariance of average symbol powers of alternative symbol sequences given by where (5) (6) (7) (8)

IEEE TRANSACTIONS ON BROADCASTING, VOL. 55, NO. 3, SEPTEMBER 2009 681 If is normalized to one, (7) can be rewritten as (9) (10) For simplicity, we assume that the average symbol power of input symbol sequence modulated with -QAM is normalized to one hereafter. If and are mutually independent, that is, two alternative symbol sequences are generated independently for a given modulation, the covariance of average symbol powers of them is zero because (9) becomes zero. However, the covariance is not zero in the conventional SLM even if the phase sequences satisfy the optimality conditions because in (10) is not one, which means that mutually independent alternative OFDM signal sequences cannot be generated by using the conventional SLM scheme for QAM modulation. Therefore, we have to design the phase sequences which change the amplitude gain to make (9) close to zero and equivalently, the fourth order joint cumulant in (6) close to zero. IV. BIT-BASED SLM SCHEMES In this section, two new bit-based SLM schemes are proposed, which change the magnitudes as well as the phases of the symbols of input symbol sequence. Also, it is shown that the covariance of average symbol powers of alternative symbol sequences in the frequency domain can be made close to zero by the proposed schemes. A. Bitwise SLM Scheme An input symbol sequence of length with -QAM can be expressed as the following binary sequence of length (11) where means the th bit of the th -QAM symbol. If a phase sequence is a binary sequence composed of with length, the alternative binary sequence is generated by multiplying the input symbol sequence in the binary form by the binary phase sequence before mapped to -QAM symbols as (12) Fig. 1. An example of partial bit inversion of Gray mapped 16-QAM constellation for PBISLM. s are mapped to -QAM symbols to generate the alternative symbol sequences, s are IFFTed, and the OFDM signal sequence with the minimum PAPR is selected for transmission, where. This SLM scheme is called bitwise SLM. is selected from -QAM symbols according to. If the binary phase sequences is any are randomly generated, the probability that -QAM symbol is, which is the same as the case of independently generated symbol sequences. Thus, we can expect that the covariance of average symbol powers of two alternative symbol sequences in the BSLM is zero. B. Partial Bit Inverted SLM Scheme In this subsection, we propose another new SLM scheme called PBISLM, where the alternative symbol sequences are generated by multiplying some preselected bits of each -QAM symbol by in the binary phase sequence,,. Let denote a subset of bit indices for -QAM symbol and be the complement set of in. The th bit of the th symbol in the binary form of the th alternative symbol sequence can be written as. (13) If is 1, the bits of corresponding to are inverted and thus is mapped to other -QAM symbol. After the alternative symbol sequences are IFFTed, the OFDM signal sequence with the minimum PAPR is selected for transmission. In the PBISLM scheme, the average power of is different from the average power of if -QAM is used, depending on the selection of the set for the given constellation mapping. From (10), we have to make as close to one as possible to have very small covariance. For some -QAM symbol mappings, we analyze the covariance of average symbol powers of

682 IEEE TRANSACTIONS ON BROADCASTING, VOL. 55, NO. 3, SEPTEMBER 2009 alternative symbol sequences in (10) in PBISLM as follows. Fig. 1 shows a Gray mapping for 16-QAM. If we use and, all bits for the th input symbol are inverted when as in Fig. 1. Assuming that, the input symbols are classified into three subsets,, and according to their powers such that with symbol power, with symbol power, and with symbol power. Then the amplitude gain of the symbol in alternative symbol sequence generated by PBISLM is. (14) If the phase sequences are randomly generated and balanced in terms of the number of 1 s and s, is also zero. Then the covariance of average symbol powers of two alternative symbol sequences in the PBISLM is calculated as Fig. 2. An example of Gray mapped 64-QAM constellation. (15) Therefore, the average symbol powers of alternative symbol sequences in the PBISLM are uncorrelated as the case of independently generated symbol sequences. Now, we consider PBISLM for the Gray mapped 64-QAM constellation given in Fig. 2. In order to compare the covariance of average symbol powers of two alternative symbol sequences, we introduce two types of PBISLM according to the selection of. For Type-I, and, and for Type-II, and. The covariance of average symbol powers of two alternative symbol sequences in Type-I is smaller than that in Type-II, e.g., for Type-I and for Type-II when, which are calculated similarly to (15). The PAPR reduction performance of these two types of PBISLM is compared through numerical analysis in the next section. The covariance of average symbol powers of two alternative symbol sequences can be made closer to zero by using different for each constellation point, but this does not guarantee and the distance property of Gray mapping at the receiver. In general, is selected such that the constellation points with the smallest power in each quadrant are mapped to the constellation points with the largest power in the opposite quadrant by bit inverting as Fig. 1 of 16-QAM and Type-I of 64-QAM. Then, in (10) can be made closest to one while keeping and the distance property of Gray mapping at the receiver. To generalize (9) for PBISLM with -QAM, we only consider the constellation points,, in the lower triangle part of the first quadrant in -QAM constellation to evaluate the symbol power. Input symbols from -QAM constellation with equal probability are classified into subsets according to their powers where the total number of groups is computed as (16) The subset index in can be replaced with, where is the constellation point in. Then, the symbol power for is given as (17) where is the distance between two closest constellation points. The probability that a symbol is in is (18) For an alternative symbol sequence generated by PBISLM, the amplitude gain of the symbol is where the power gain is (19) (20) If the phase sequences are orthogonal to each other and balanced in terms of the number of 1 s and 1 s, the covari-

IEEE TRANSACTIONS ON BROADCASTING, VOL. 55, NO. 3, SEPTEMBER 2009 683 and the size of side information in BSLM and PBISLM is the same as that in the conventional SLM for the same. C. Shaping Gain of the Proposed Schemes and Their PAPR Since the proposed schemes modify the average power of alternative symbol sequences, the average power of the OFDM signal sequence selected for transmission can be different from that of the input symbol sequence. By using the proposed schemes, the average power of the OFDM signal sequence can be reduced, which is regarded as a power shaping scheme [20]. The shaping gain is the gain due to the average power reduction defined as (23) Fig. 3. Comparison of the covariance of average symbol powers of two alternative symbol sequences for various SLM schemes. ance of average symbol powers of two alternative symbol sequences is given as where and. In [21] and [22], PAPR reduction schemes using trellis shaping or MMSE are proposed, which achieve shaping gain as well as PAPR reduction. In [21], the SNR for AWGN channel is derived by considering the shaping gain as where the normalized symbol power is (21) (22) The alternative symbol sequences are generated by multiplying an input symbol sequence with a phase sequence in the conventional SLM. But in practice, this multiplication is performed by changing the constellation points of an input symbol sequence to another constellation points according to a phase sequence. BSLM and PBISLM also change the constellation points of an input symbol sequence to different constellation points according to a binary phase sequence. The rest of the procedures in the proposed schemes are the same as those in the conventional SLM. Therefore, the complexity of the proposed scheme is also the same as that of the conventional SLM. In BSLM, we may need more memories or registers to store the phase sequences. But the elements of phase sequences are binary values and the amount of the increased memory or register is negligible. It is clear that BSLM and PBISLM also preserve the distance property of Gray mapping after removing the binary phase sequence at the receiver similarly as the conventional SLM scheme. Therefore, there is no BER degradation in both SLM schemes assuming that the receiver knows perfect side information. In the proposed schemes, side information should be sent along with the data similarly to the conventional SLM scheme (24) where for -QAM, is the transmit energy per information bit, and is the one-sided power spectral density of an AWGN process. in the second term corresponds to the rate of loss due to SLM scheme, that is, the amount of side information which is the same for the proposed SLM schemes and the conventional SLM scheme. in the third term is the shaping gain defined in (23) which is one for the conventional SLM and more than one for the proposed SLM schemes. In the next section, it will be shown that two proposed schemes also have the shaping gain in addition to PAPR reduction through numerical analysis. Thus PAPR in (2) should be normalized by power shaping gain as V. SIMULATION RESULTS (25) In this section, we compare the covariance of average symbol powers of two alternative symbol sequences for the conventional SLM and the proposed schemes and also compare the PAPR reduction performance of them for. The rows of cyclic Hadamard matrix are used for phase sequences [16] and the all-1 sequence is used for to include the original input symbol sequence among the alternative symbol sequences. Fig. 3 compares the covariance of average symbol powers of two alternative symbol sequences for, 128, 256 and 512 when 16-QAM and 64-QAM are used. The conventional SLM scheme has the largest covariance and the covariance for 64-QAM is larger than that for 16-QAM. In the case of PBISLM

684 IEEE TRANSACTIONS ON BROADCASTING, VOL. 55, NO. 3, SEPTEMBER 2009 Fig. 5. Shaping gain of BSLM and PBISLM with N =64and 256 for U = 8; 16, and 32. Fig. 4. Comparison of PAPR reduction performance of the conventional SLM, proposed schemes, and theoretical CCDF of (3) for U =8: (a) N =64; (b) N =256. for 64-QAM, the covariance in Type-I is smaller than that in Type-II. Although the covariance does not become zero for the proposed schemes for smaller, it is clear the covariance in the proposed schemes is closer to zero than that of the conventional SLM for any and. Fig. 4 compares the PAPR reduction performance of various SLM schemes for and 256. Note that oversampling is not performed when we compare the results with the theoretical CCDF in (3). The conventional SLM scheme shows the worst performance, especially for 64-QAM, but the performance gap with other schemes decreases as increases. The PAPR reduction performance of PBISLM with Type-II is slightly worse than that of PBISLM with Type-I for. The PAPR reduction performance of BSLM and PBISLM with Type-I is almost identical to the theoretical CCDF curves in (3). Note that the PAPR reduction performance shows similar tendency as the covariance given in Fig. 3. Now, the shaping gain is considered to evaluate the PAPR reduction performance. Fig. 5 shows the shaping gain of BSLM and PBISLM (Type-I for 64-QAM) through the numerical analysis. The shaping gain increases as increases, decreases, Fig. 6. Comparison of PAPR reduction performance of the conventional SLM and proposed schemes for U =16: (a) N =64; (b) N = 256. or the modulation order increases. Two proposed schemes have the same shaping gain for 16-QAM but BSLM has slightly

IEEE TRANSACTIONS ON BROADCASTING, VOL. 55, NO. 3, SEPTEMBER 2009 685 bigger shaping gain than PBISLM for 64-QAM. Since the required power is reduced to transmit the same amount of information, SNR gain can be obtained from the shaping gain as in (24) and therefore BER performance is improved assuming that the distance property of QAM at the receiver is preserved with perfect side information. Fig. 6 compares the reduction performance in (25) of BSLM, PBISLM (Type-I for 64-QAM), and conventional SLM for and 256 when 16-QAM and 64-QAM,, and four times oversampling are used. It is clear that the reduction performance of the proposed schemes is still better than that of the conventional SLM. VI. CONCLUSION We proposed two new bit-based SLM schemes for PAPR reduction in QAM modulated OFDM signals, called BSLM and PBISLM. The proposed schemes modify the magnitude as well as the phase by applying binary phase sequence to the input symbol sequence in the binary form to generate alternative OFDM signal sequences. The proposed schemes do not increase the computational complexity and the amount of side information compared to the conventional SLM. The improvement in the PAPR reduction performance of two proposed schemes increases as increases and decreases for QAM modulated OFDM signals. Simulation results show that the proposed schemes have better PAPR reduction performance than the conventional SLM and converge to the theoretical CCDF. Furthermore, the proposed schemes have shaping gain which can improve BER performance. Especially, BSLM does not depend on the constellation mapping scheme and has better PAPR reduction performance and more shaping gain than other SLM schemes. REFERENCES [1] R. O neal and L. N. Lopes, Envelope variation and spectral splatter in clipped multicarrier signals, in Proc. PIMRC 95, Sep. 1995, pp. 71 75. [2] D. Wulich and L. Goldfeld, Reduction of peak factor in orthogonal multicarrier modulation by amplitude limiting and coding, IEEE Trans. Commun., vol. 47, no. 1, pp. 18 21, Jan. 1999. [3] X. Wang, T. T. Tjhung, and C. S. Ng, Reduction of peak-to-average power ratio of OFDM system using a companding technique, IEEE Trans. Broadcast., vol. 45, no. 3, pp. 303 307, Sep. 1999. [4] T. Jiang and G. Zhu, Nonlinear companding transform for reducing peak-to-average power ratio of OFDM signals, IEEE Trans. Broadcast., vol. 50, no. 3, pp. 342 346, Sep. 2004. [5] S. A. Aburakhia, E. F. Badram, and A. E. Mohamed, Linear companding transform for the reduction of peak-to-average power ratio of OFDM signals, IEEE Trans. Broadcast., vol. 55, no. 1, pp. 155 160, Mar. 2009. [6] J. Tellado and J. M. Cioffi, Multicarrier Modulation with Low PAR, Application to DSL and Wireless. Norwell, MA: Kluwer Academic Publisher, 2000. [7] B. S. Krongold and D. L. Jones, PAR reduction in OFDM via active constellation extension, IEEE Trans. Broadcast., vol. 49, no. 3, pp. 258 268, Sep. 2002. [8] R. W. Bäuml, R. F. H. Fischer, and J. B. Huber, Reducing the peak-toaverage power ratio of multicarrier modulation by selected mapping, Electron. Lett., vol. 32, no. 22, pp. 2056 2057, Oct. 1996. [9] S. H. Müller, R. W. Bäuml, R. F. H. Fischer, and J. B. Huber, OFDM with reduced peak-to-average power ratio by multiple signal representation, Ann. Telecommun., vol. 52, no. 1-2, pp. 58 67, Feb. 1997. [10] D.-W. Lim, S.-J. Heo, J.-S. No, and H. Chung, A new PTS OFDM scheme with low complexity for PAPR reduction, IEEE Trans. Broadcast., vol. 52, no. 1, pp. 77 82, Mar. 2006. [11] S.-J. Heo, H.-S. Noh, J.-S. No, and D.-J. Shin, A modified SLM scheme with low complexity for PAPR reduction of OFDM systems, IEEE Trans. Broadcast., vol. 53, no. 4, pp. 804 808, Dec. 2007. [12] T. Jiang, W. Xiang, P. C. Richardson, J. Guo, and G. Zhu, PAPR reduction of OFDM signals using partial transmit sequences with low computational complexity, IEEE Trans. Broadcast., vol. 53, no. 3, pp. 719 724, Sep. 2007. [13] L. Yang, K. K. Soo, Y. M. Siu, and S. Q. Li, A low complexity selected mapping scheme by use of time domain sequence superposition technique for PAPR reduction in OFDM system, IEEE Trans. Broadcast., vol. 54, no. 4, pp. 821 824, Dec. 2008. [14] A. D. S. Jayalath and C. Tellambura, Reducing the peak-to-average power ratio of orthogonal frequency division multiplexing signal through bit or symbol interleaving, IEE Electron. Lett., vol. 36, no. 13, pp. 1161 1163, Jun. 2000. [15] M. Breiling, S. H. Müller, and J. B. Huber, SLM peak-power reduction without explicit side information, IEEE Commun. Lett., vol. 5, no. 6, pp. 239 241, Jun. 2001. [16] D.-W. Lim, S.-J. Heo, and J.-S. No, On the phase sequence set of SLM OFDM scheme for a crest factor reduction, IEEE Trans. Signal Process., vol. 54, no. 5, pp. 1931 1935, May 2006. [17] G. T. Zhou and L. Peng, Optimality condition for selected mapping in OFDM, IEEE Trans. Signal Process., vol. 54, no. 8, pp. 3159 3165, Aug. 2006. [18] D. R. Brillinger, Time Series, Data Analysis and Theory. San Francisco, CA: Holden-Day, 1981. [19] S.-L. J. Hu, Probabilistic independence and joint cumulants, J. of Engin. Mech., vol. 117, no. 3, pp. 640 652, Mar. 1991. [20] G. D. Forney, Trellis shaping, IEEE Trans. Inf. Theory, vol. 38, no. 2, pp. 281 300, Mar. 1992. [21] H. Ochiai, A novel trellis-shaping design with both peak and average power reduction for OFDM systems, IEEE Trans. Commun., vol. 52, no. 11, pp. 1916 1926, Nov. 2004. [22] F. Kohandani and A. K. Khandani, A new algorithm for peak/average power reduction in OFDM systems, IEEE Trans. Broadcast., vol. 54, no. 1, pp. 159 165, Mar. 2008.